A Performance Study on Emotion Models Detection Accuracy in a Pandemic Environment

No Thumbnail Available
Saravanan P.
Ravindran S.
Weng L.Y.
Mohamed Sahari K.S.B.
Anuar A.B.
Abdul Jalal M.F.B.
Mohamad Rafaai Z.F.B.
Raventhran P.N.A./P.
Mohd Radzi H.
Yussof S.
Journal Title
Journal ISSN
Volume Title
Springer Science and Business Media Deutschland GmbH
Research Projects
Organizational Units
Journal Issue
This paper studies emotion detection using deep learning on the prevalent usage of face masks in the Covid-19 pandemic. Internet repository data Karolinska Directed Emotional Faces (KDEF) [1] was used as a base database, in which it was segmented into different portions of the face, such as forehead patch, eye patch, and skin patch to be representing segments of the face covered or exposed by the mask were transfer learned to an Inception v3 model. Results show that the full-face model had the highest accuracy 74.68% followed by the skin patch (area occluded by the mask) 65.09%. The models trained on full-face were then used to inference the different face segments/patches that showed poor inferencing results. However, certain emotions are more distinct around the eye region. Therefore, this paper concludes that upper segmented faces result in higher accuracy for training models over full faces, yet future research needs to be done on additional occlusion near the eye section. � 2021, Springer Nature Switzerland AG.
Computers; Covid19; Deep learning; Detection accuracy; Emotion; Emotion models; Full faces; High-accuracy; Pandemic; Performance study; Skin patch; Deep learning